An optimal scheme for differentiating a variety of fruits for real-time prediction using Convolutional Neural Network by comparing with Decision Tree to increase the Accuracy, Precision and Sensitivity

Aim: The objective of this study is to implement deep learning technique based on innovative classification using convolutional neural networks algorithm to improve the detection of disease in fresh fruit compared with the Decision Tree algorithm to increase accuracy, precision, and sensitivity. Materials and Methods: A total of 2500 samples were collected from kaggle’s various fruit images. A training dataset of 1750 (70%) and a test dataset of 750 (30%) were generated from these samples. To measure the efficiency of the Convolutional neural network algorithm, accuracy, precision, and sensitivity values were measured. For each sample, 10 samples were collected for analysis. The significance p<0.005 mean accuracy, precision, and sensitivity shows the performance of fruit quality detection. The study parameters are found as alpha = 0.02 and G power = 0.8. Results: On the two classes under consideration, an independent samples T-test was performed. CNN has achieved accuracy, precision, and sensitivity of 97.5%, 95.1% and 99.4% respectively compared to 87.7%, 90.8%, and 91.2% by the Decision Tree algorithm. Conclusion: In this study it was found that the CNN algorithm appears to be significantly better than the Decision Tree algorithm in fruit quality detection whether it was healthy or diseased of the datasets considered.